Intent is a hypothesis that when implementation cost collapses, the bottleneck moves upstream to specification clarity — and that ticket-based coordination tools optimize for the old bottleneck. We are testing this hypothesis in public, with the methodology running on itself, and inviting you to pressure-test it.
The loop. Four phases, continuous motion, no sprint boundaries. Each transition emits an event. Multi-speed — some loops take minutes, some take weeks.
Staff+ engineers on teams of 2–7 using Claude Code daily.
When AI agents can implement well-specified work in hours instead of weeks, the expensive part of software delivery moves from writing code to getting the specification right. Teams that keep optimizing for the old constraint — coordinating human labor through tickets, sprints, and story points — are building overhead around the wrong bottleneck.
This claim might be wrong. We have one external practitioner who described the pattern independently before we named it, plus three independent 2026 convergences from published work: Marty Cagan's context-engineering coaching stack, Block's BuilderBot agent fleet, and the MobAI framing of team coherence (Joe Justice, Michael Göthe). Convergence is not discovery: none of those test Intent itself, and the 194 internal signals from building Intent still outweigh every external datum. Structured discovery with the target user remains the open gate, held at the irreversible boundary of public generalization claims (gate the irreversible, select the reversible); we'll publish whatever it finds, including disconfirmation.
If you're a staff+ engineer on a small team running Claude Code every day, you are already solving the spec-clarity problem somehow. Today that is almost certainly one of these:
CLAUDE.md / AGENTS.md — fine until the context outgrows one file and nobody can say why a rule is there..intent/ or docs/ scaffold from scratch — every team reinvents signals, specs, trust levels, and observability, and each reinvention is invisible to the next.Intent is the alternative to reinventing that scaffold yourself. It's a file-native, git-tracked operating model — Notice → Spec → Execute → Observe — with the signal schema, trust math, and observability already shaped, so you adopt a discipline instead of authoring one from a blank file. If what you actually need is the bird's-eye view of how a whole portfolio of products composes — not the operating model itself — that lives on Parallax, not here.
Intent is a synthesis, not a novel method. The loop — Notice → Spec → Execute → Observe — is a rename of patterns our field has known for decades. We credit the ancestors explicitly because the credibility travels with them.
Continuous discovery is Teresa Torres. Story mapping is Jeff Patton. The product operating model is Marty Cagan and Melissa Perri. Outcomes over outputs is Josh Seiden. OODA is John Boyd (1976). PDCA is Deming. Build-Measure-Learn is Eric Ries. Psychological safety is Amy Edmondson — and it is a prerequisite Intent is still working to earn. GIST planning and evidence-based decision rigor are Itamar Gilad. Category positioning is April Dunford — and her feedback shaped this page.
The site is organized as four zones of progressively deeper content. Each zone is honestly labeled about what's tablestakes, what's evolutionary, and what's still an open question.
The further you go, the more you're looking at work-in-progress. The Hypothesis zone is written. The System and Build zones carry the full depth pages (unified in the 2026-06-05 overhaul). The Proof zone is a live ledger.
Intent is not shipped product. It is an active research hypothesis with a working prototype. The architecture has known gaps (we published the panel review that flagged them). The discovery base is still thin: one external practitioner voice, plus three published-work convergences logged in 2026 (Cagan, Block, MobAI) that corroborate the problem shape without testing the method. Structured discovery with the target user is the next gate, kept at the irreversible boundary of outward generalization claims, not on the building itself, and not yet run. The methodology is missing a psychological safety contract v2 and a change management rollout playbook. We are building those in public, and you can read them as we draft them.
Intent is not for every team. It requires blue-green deployment or equivalent rollback capability, feature flag change management, significant automated testing, and visible change reporting. Without that infrastructure, the trust formula is fictional and L3/L4 agent autonomy is unsafe. It also requires a culture that values speaking up — if your organization punishes honest mistakes, Intent's mechanisms will become a surveillance layer. See "when NOT to adopt Intent".
Intent asks teams to release things. Sprint rhythm. Standups as social presence. Story points. The identity of "the person who runs our process." These are real losses for the humans affected, and we are explicit about them. See what Intent asks you to release, and who loses power when Intent is adopted.
We'll know the hypothesis is wrong if: after 10 external discovery interviews, fewer than 7 participants describe the spec-clarity bottleneck in their own words, unprompted. We'll publish the result whether it confirms or disconfirms.
This site is about the operating model — the system, not the showroom. Intent is the operating model that forged a working portfolio of products at The Parlor; on that dashboard every product points back here as "the operating model that built me." Two systems sit behind the portfolio's veil: Intent (this — how the work runs) and Parallax (the overhead view of how the whole stack composes). You don't need the worldview to use Intent — it's here when you want it.
An honest doorway, not a catalog — the product list lives on the portfolio surface, the worldview on Parallax. This page stays about the discipline.